Artificial intelligence is moving deeper into pharmaceutical research, where finding and improving potential medicines can require years of experiments and extensive scientific analysis. Anthropic has reported new results from its work with six life-sciences organizations, describing how its Claude AI models have supported tasks ranging from antibody design to drug-manufacturing documentation and clinical-trial planning.
In one of the most striking examples, Anthropic says Claude helped biotechnology company Divergence Labs generate approximately 35,000 antibody designs in about a month for an experimental approach to treating food allergies. The results illustrate how AI can help researchers explore large numbers of possible molecular designs, although laboratory testing and further development are still required before any candidate could become a medicine.
How AI Helped Generate Thousands of Antibody Designs
Antibodies are proteins that can recognize and bind to specific targets in the body. Designing therapeutic antibodies requires researchers to balance several properties, including target binding, stability, safety and the ability to manufacture a potential treatment reliably. AI tools can help explore molecular designs, but scientists must still determine which options work in laboratory experiments and whether they are suitable for further development.
According to Anthropic, Divergence Labs is investigating a potential treatment for food allergies that aims to detach immunoglobulin E, or IgE, from immune cells. IgE plays a central role in many allergic reactions. The company's approach involves designing an antibody with two functional parts: one intended to bind to IgE and another intended to help remove it from immune cells.
Anthropic reports that Claude generated roughly 35,000 antibody designs over about a month. In early testing, Divergence identified three designs that bound to IgE and 20 that could dislodge it from immune cells. Three of the latter group appeared to outperform a standard medication on that specific experimental measure. Researchers still need to combine the desired functions into a potential drug and conduct further studies.
AI Also Helped Improve Experimental Cancer-Related Molecules
Another project involved Generate Biomedicines, which is exploring ways to adapt CAR T-cell therapy for solid tumours. CAR T-cell treatments modify immune cells so they can recognize and attack cancer cells, but extending the approach to solid tumours presents substantial scientific challenges.
Anthropic says six Generate scientists used Claude Code to propose designs for a shorter section of the genetic instructions used in the therapy. One proposed design was about 70% shorter than the standard version. In laboratory tests reported by the companies, it enabled immune cells to produce roughly ten times as many receptors, killed tumour cells more efficiently and helped the cells remain active for longer. These findings relate to an experimental research component, not a proven cancer treatment.
Anthropic also described work with BigHat Biosciences, where AI-supported suggestions were used to improve experimental antibody fragments. The company reports that the best versions in the project bound more than ten times more tightly to their target than the originals. Other measured improvements included greater heat stability and slower clumping, properties that can affect whether a molecule is practical to develop into a medicine.
AI Is Being Used Beyond Molecular Design
Drug development involves far more than discovering a promising molecule. Researchers must assess potential targets, refine candidate compounds, prepare manufacturing processes, review regulatory documentation and plan clinical trials. Anthropic argues that AI could help across this wider workflow, rather than being used only to generate initial molecular designs.
In a project with Enveda, a biotechnology company working on drug discovery, an AI-supported review system examined 1,061 pages of manufacturing documentation. Anthropic says the system identified all 11 issues previously found by human reviewers and flagged seven additional issues. The company also reported an 89% reduction in human review time during the test, from more than 25 hours to less than three. These results are specific to the reported document-review exercise and should not be interpreted as a general guarantee of error-free AI review.
Anthropic also said Bristol Myers Squibb has used Claude to analyse an early-stage drug's clinical-trial plans and assess the likelihood of trial success. The companies intend to expand this work in 2027. AI-generated analysis can support scientific decisions, but clinical trials still require rigorous protocols, human oversight and evidence from patients.
Why These Results Matter for Pharmaceutical Research
One of the biggest challenges in drug development is deciding which possibilities deserve expensive and time-consuming experiments. AI systems can help researchers explore a broader set of designs, organize complex data and identify patterns that might otherwise take longer to investigate. When connected to laboratory systems, these models may also help researchers iterate between proposed designs and experimental results.
The potential benefit is not that an AI model can independently guarantee a cure. Instead, AI could help scientists test ideas more efficiently, reduce some administrative burdens and prioritize promising research directions. The impact will depend on whether initial results can be reproduced, whether candidate molecules prove safe and effective, and whether the technology delivers measurable benefits across different programmes.
Important Limits: No Approved Cures Yet
Anthropic explicitly describes much of this work as being at an early, preclinical stage, meaning most of the reported drug candidates have not yet reached human testing. A promising result in a laboratory experiment does not establish that a treatment will be safe or effective in people. Candidates can fail during further testing because of toxicity, weak effectiveness, manufacturing problems or other scientific challenges.
The reported numbers also come from Anthropic and its research partners. Independent replication and further technical evidence will be important for assessing how broadly the findings apply. The work is promising as an example of AI-assisted research, but it should not be confused with a medical breakthrough that is already available to patients.
What Comes Next for AI Drug Discovery?
Anthropic says it plans to expand its work with biotechnology companies and deepen the role of Claude across the drug-development process. The company has also described its own research efforts involving rare childhood forms of muscular dystrophy. Any future clinical milestones will depend on the results of further experiments, regulatory requirements and trials involving human participants.
The wider significance is that AI is being tested across multiple stages of pharmaceutical research, from designing potential molecules to reviewing documents and supporting trial planning. If these workflows continue to produce reproducible improvements, they could change how research teams explore and evaluate drug candidates. For now, the evidence points to an emerging research tool with meaningful potential, but substantial scientific work remains before the reported projects can produce approved medicines.
